Estimating parameters of stochastic radio channel models based on new measurement data is an arduous task usually involving multiple steps such as multipath extraction and clustering. We propose two different machine learning methods, one based on approximate Bayesian computation (ABC) and the other on deep learning, for fitting stochastic channel models to data directly. The proposed methods make use of easy-to-compute summary statistics of measured data instead of relying on extracted multipath components. Moreover, the need for post-processing of the extracted multipath components is omitted. Taking the polarimetric propagation graph model as an example stochastic model, we present relevant summaries and evaluate the performance of the proposed methods on simulated and measured data. We find that the methods are able to learn the parameters of the model accurately in simulations. Applying the methods on 60 GHz indoor measurement data yields parameter estimates that generate averaged power delay profile from the model that fits the data.

Calibration of stochastic radio channel models is the process of fitting the parameters of a model such that it generates synthetic data similar to the measurements. The traditional calibration approach involves, first, extracting the multipath components, then, grouping them into clusters, and finally, estimating the model parameters.In this paper, we propose to use approximate Bayesian computation (ABC) to calibrate stochastic channel models so as to bypass the need for multipath extraction and clustering. We apply the ABC method to calibrate the wellknown Saleh-Valenzuela model and show its performance in simulations and using measured data. We find that the Saleh-Valenzuela model can be calibrated directly without the need for multipath extraction or clustering.

This paper proposes a method to infer on the parameters of a stochastic channel model from observations of temporal moments without multipath extraction. The distribution of the temporal moments is approximated to be Gaussian, and sampling is carried out from the approximate posterior. The temporal moments are found to be informative about the model parameters, as the parameters can be recovered from the samples.

Stochastic channel models are usually calibrated after extracting the parameters of the multipath components from measurements. This paper proposes a method to infer on the underlying parameters of a stochastic multipath model, in particular the Turin model, without resolving the multipath components. Channel measurements are summarised into temporal moments instead of the multipath parameters. The parameters of the stochastic model are then estimated from the observations of temporal moments using a method of moments approach. The estimator is tested on real data obtained from in-room channel measurements. It is concluded that calibration of stochastic models can be done without multipath extraction, and that temporal moments are informative summary statistics about the model parameters.

No abstract